OpenAI has started showing multiple products in a single ad placement inside ChatGPT, a product carousel that sits inside a conversation and pulls straight from a retailer's product feed, the same way Google Shopping works. It landed last week, about three months after OpenAI first let advertisers upload bulk product catalogues.

Right now, each carousel only shows products from one retailer.

OpenAI's own systems decide whether a query gets a single product ad or a full carousel, the advertiser doesn't get to choose, and there's nothing in how it's built that stops a future version showing competing retailers side by side in the same carousel.

Once OpenAI enables this, you're looking at a real auction for shelf space inside a ChatGPT answer.

I'm very excited about this one. It's the clearest sign yet that OpenAI wants ChatGPT to be a shopping surface (as they slowly build towards this following the rushed Instant Checkout failure), not just an answer engine.

I ran "best maternity jeans" through Obsero's prompt capture on ChatGPT twice this week. Same result both times, Quince is the only brand in the carousel, five products showing, different items each run but the same brand holding every slot.

Quince features for every slot in that carousel, regardless of whether it was ever the brand ChatGPT would have recommended organically.

That's exactly what we built the alerting for. When a tracked prompt returns a product carousel, Obsero flags it the same way it shows a map pack or a local business panel, so you know when paid placement is turning up next to, or instead of, your organic mention.

That matters because organic visibility and paid placement are now playing in the same window - not too dissimilar to when I started in SEO back in 2006 (with PPC clearly labelled as sponsored).

You can be the best-cited answer for a category and still lose the real estate at the bottom of the screen to whoever's bidding on that carousel. It is worth watching who shows up here over the next few months, and worth deciding early whether it's an area you want to play in.

Big news out of Google DeepMind last week. Demis Hassabis is stepping down as CEO of Google DeepMind to become chairman, moving up to chief scientist of Alphabet as a whole, while still running Isomorphic Labs, his drug discovery arm. Koray Kavukcuoglu, the outgoing CTO, takes over day to day, reporting straight to Sundar Pichai.

At the same time, Jeff Dean, Google's chief scientist for the past 27 years, is leaving too. He's founding Discovery Loop, an independent public benefit corporation, with Google staying on as investor and cloud provider. Three more senior researchers are going with him, Sanjay Ghemawat, Oriol Vinyals and Quoc Le, one of the co-founders of Google Brain.

That is a lot of talent exiting the door at Google over the past three months. Noam Shazeer and John Jumper recently left the business.

Hassabis says the timing is about progress towards AGI, handing over operational responsibility to focus on the bigger picture. Dean's framing is more pointed. He's said Discovery Loop, built as a public benefit corporation, can make decisions that aren't necessarily in its purest financial interest, free from Google's shareholders.

Earlier in the year, I read The Infinity Machine, Sebastian Mallaby's book on Hassabis and it gives a fascinating insight into the brilliance but yet naivety of the DeepMind founder.

Hassabis is a genius, that's not in question. Chess prodigy as a teenager, working under the difficult Peter Molyneux at Bullfrog on Theme Park, one of the best-selling games of the 90s, before he'd even finished his Cambridge degree. All from a boy that grew up about five miles away from where I lived - he truly is a unique talent.

What I discovered is how close DeepMind came to going somewhere other than Google. Elon Musk was an early investor and reportedly tried to block the Google deal once it was on the table. Zuckerberg was in the running too, and Hassabis is said to have used Meta's interest as leverage to get better terms out of Google. Google won in 2014.

But the tension the book keeps coming back to is this. Hassabis built DeepMind to create AGI that benefits humanity, not to sell more ads. That's a mission, not a business model, and I don't think it's ever sat comfortably inside a company that makes the overwhelming majority of its money from search and ads.

You can see it play out. Google Brain and DeepMind merged in 2023, with Hassabis ending up in charge of the combined unit. Sergey Brin coming out of retirement to get hands-on with Gemini strategy alongside Hassabis is another version of the same tension, a co-founder (and also genius) with obvious commercial instincts working next to someone whose stated goal has always been bigger than the product roadmap.

The clearest example: a former Google engineer, Thibault Sottiaux, said publicly this week that Google had a ChatGPT-style product ready a full year before OpenAI launched, codenamed LMChat. It didn't ship. His words, Google was too nervous to release it, and DeepMind was blocked from shipping anything that could disrupt search.

That is insane. Google was arguably the most capable AI lab on Earth for years and still let a smaller, hungrier competitor eat its lunch, because shipping the product it already had would have put its own advertising business at risk.

The obvious one is more ads inside Gemini and AI Mode. Once the AGI-for-humanity voice at the top of DeepMind steps back into a chairman role, the commercial instincts that Brin and Pichai represent get a freer hand.

It's also clearly a power struggle, not just a scheduling coincidence, worth watching who ends up with real authority over Gemini's roadmap versus who just keeps the title.

And there's a real risk Gemini 3.5 undershoots. It wasn't long ago that Sam Altman was calling a code red at OpenAI after Gemini 3 landed in November, that's how fast this race can flip. Early reporting on 3.5 points to it lagging competitors by months, with senior researchers already leaving for Anthropic and OpenAI.

I won't get into the internal politics or the model capability arguments in depth here, the semianalysis team cover that better than I will. But if you want the human story behind where Google and Hassabis find themselves today, I highly recommend buying the book.

As always, thanks for subscribing. I do love to hear your feedback, so please reply to this email and let me know what you thought of this edition.

Until next week,

Andy